Forty years of Landsat and Sentinel-2 retrospective analysis reveal a species-differentiated dieback signal in the Atlas forests of Béni Mellal-Khénifra (Morocco)

Abstract

Long-term, spatially explicit reconstructions of forest dieback are scarce for the southern African rim of the Mediterranean basin.This study presents an open, reproducible retrospective analysis of natural forest dieback in the Béni Mellal-Khénifra region of Morocco (936,437 ha) over 1984-2024.Six strata of the Second National Forest Inventory dominated by Cedrus atlantica, Quercus rotundifolia and Quercus suber were sampled through a stratified random design (2.901 Landsat pixels; 2.933 Sentinel-2 pixels).Per-pixel NDVI series were processed with a BFAST-like workflow combining STL decomposition and PELT change-point detection, and complemented by a recent-decline detector to rescue near-edge signals.Species-level NDVI anomalies were cross-correlated with SPEI at 3-, 6-and 12-month scales and with vapour-pressure-deficit anomalies, with autocorrelation-corrected significance testing.A cumulated dieback signal of approximately 187,700 ha (20.1% of the regional forest cover; bootstrap 95% CI 155,500-221,600 ha) was identified.Cork oak showed the highest relative rate (41.4% of pure stands) and holm oak the largest absolute affected area (142,591 ha).SPEI-12 emerged as the strongest hydroclimatic correlate, particularly for cork oak (r = +0.364at lag 0), while VPD anomalies co-varied with concurrent NDVI depressions in all three species.Sentinel-2 detected +18.3 to +26.3 percentage points more Atlas cedar mortality than Landsat, whereas Landsat's longer baseline captured cork-oak historical pulses.The analysis contributes a reproducible baseline for regional forest management and a transferable analytical template for other Maghreb regions where dieback dynamics remain poorly documented.